The existing literature acknowledges that a mismatch between the experimenter’s and the subjects’ models of an experimental task can adversely affect the interpretation of data from laboratory experiments. We discuss why the two common experimental designs (between-subjects and within-subjects) used to conduct experiments may fail to sufficiently account for this concern. An alternative design for laboratory experiments is proposed which may alleviate this concern especially in studies of social preferences. The proposed design is used to answer some questions that have attracted continued attention in the literature on social preferences in general and reciprocity in particular.